Articles
Harold Cohen and AARON
n Harold Cohen (1928–2016) was a
computer art pioneer and the developer
of AARON, one of the most creative AI
programs in daily use. This memoriam,
written by his son (and AI scientist)
Paul, describes Harold Cohen’s work
and contributions to the AI and art
communities. — ed.
Copyright © 2016, Association for the Advancement of Artificial Intelligence. All rights reserved. ISSN 0738-4602
Paul Cohen
arold Cohen was the author of AARON, perhaps the
H longest-lived and certainly the most creative artificial
intelligence program in daily use. Cohen’s death, on
April 27, 2016, ended a lengthy partnership between an artist
and an AI. At times, AARON was quite autonomous of
Cohen, responsible for the composition, coloring, and other
aspects of a work; more recently, AARON served Cohen by
making drawings that he would develop into paintings.
Cohen was born in 1928, studied painting at the Slade
School of Fine Arts in London, and later taught at the Slade
as well as Camberwell, Nottingham, and other art schools.
He represented Great Britain at major international festivals
during the 1960s, including the Venice Biennale, Documen-
ta 3, and the Paris Biennale. He showed widely and success-
fully. Then, in 1968, he left the familiar and rewarding Lon-
don scene to join the Visual Arts Department at the
University of California, San Diego. By 1971, Cohen had
taught himself to program a computer and exhibited com-
puter-generated art at the Fall Joint Computer Conference.
Thereafter, but for his notebook drawings and the occasion-
al birthday card, every image from Cohen’s studio in some
way involved the AARON program.
Cohen’s fundamental question was, “What makes an
image evocative?” Can algorithms produce evocative images
not rarely and accidentally but consistently? In the early
1970s, Cohen realized that some aspects of marks — open
and closed shapes, implied figure and ground, lines that look
like they are drawn by hand rather than, say, Bezier curves —
gave the impression of intentionality (figure 1). Images from
this period look a bit like abstract Native American petro-
WINTER 2016 63
Articles
Figure 1. An Early Drawing by AARON, 1974.
Colored in pencil by Harold Cohen.
glyphs, or abstract children’s drawings, two early
influences on the development of AARON. The
images were black and white. Cohen colored them by
hand.
For the next two decades, Cohen worked on algo-
rithms for color and composition. In 1995, he built
and exhibited a painting machine at the Boston
Computer Museum. Cohen was a talented engineer:
His machine would compose images of people in
rooms, then draw them, mix its own dyes, and color
the drawings (figure 2). This exhibition turned out to
be the apex of AARON’s career as an autonomous rep-
resentational artist. Representational painting was
evocative, but not in interesting ways, and although
Cohen loved to interact with gallery audiences, he
worried that the spectacle of the painting machine
detracted from the art itself. Gradually, Cohen and
AARON returned to abstraction and simplicity,
although their path wasn’t direct. For roughly a
64 AI MAGAZINE
decade, AARON produced somewhat representation-
al images of astonishing complexity and breathtak-
ing color that Cohen regarded as superior to his own
(figure 3). But Cohen’s question had always been,
“What are the minimum conditions under which a
set of marks functions as an image?” and his work
since about 2010 is comparatively minimalist.
The early AARON was developed at the Stanford AI
Laboratory, where Cohen was a sabbatical guest of
professor Edward A. Feigenbaum. AARON was an
expert system that encoded Cohen’s knowledge about
composition and color. Even in these early days there
were hints — such as the freehand line algorithm —
that art might be made by algorithms that have little
in common with artists’ expertise. Composition and
color were eventually realized by startlingly simple
algorithms. Cohen said that AARON was a distinctly
different kind of intelligence from his own and he
puzzled over its surprising success.
 Articles

Figure 2. A Close-Up of One of Harold Cohen’s Painting Machines.

On the left is an effector that grips, raises, and lowers brushes, attached to the beam of an x-y plotter. The paint cup on the right is also
attached to the beam. To color an area, AARON would select a brush of appropriate size and mix dyes in the cup.

Cohen quipped that he would be the first artist in

history to have a posthumous exhibition of new
work, though he realized that the joke hinged on the
word “new.” Each of AARON’s images is both unique
and instantly recognizable as originating in a partic-
ular version of the code. Cohen toyed with the idea
of making AARON modify itself, but in fact he
reduced AARON’s autonomy. Why was this? He had
little faith in machine learning, he believed that only
an artist could make AARON, he wanted to retain
control of AARON’s development; most importantly,
AARON was how he made art. The final version of
AARON draws images that were extremely challeng-
ing for Cohen to paint: few marks, lots of white
space, no hints about color (figure 4). As an artist
who never erased, Cohen would think for hours and
days about how to paint these drawings. AARON
became not an artist in its own right but a poser of
problems to an artist. And Cohen, having demon-
strated that AARON could function as an artist in the Figure 3. Harold Cohen with an Installation at the San Diego Museum
world’s galleries, museums, and science centers — the of Contemporary Art, 2007.
Tate, the Stedelijk, the San Francisco Museum of Art, The drawing and coloring were done entirely by AARON. The painting was
the Ontario Science Center, and many others — printed on a wide-format inkjet printer in seven panels.

WINTER 2016 65
Articles

Figure 4. “Wakl in Rio.”

Drawing by AARON, coloring by Harold Cohen, 2015. Cohen painted some of these drawings in oil paint, but eventually built yet anoth-
er drawing machine: a six-foot display with touch sensors and custom software to allow Cohen to mix colors and paint with his fingers
directly on the display. The finished images were printed on canvas.

spent his final years doing what he liked best: using
AARON to think about color and composition and
the conditions under which marks function as
images.
AARON’s images and Cohen’s essays and videos
can be viewed at www.aaronshome.com. The book
Aaron’s Code by Pamela McCorduck documents the
first phases of AARON’s career.
Paul Cohen is an AI researcher and a professor at the Uni-
versity of Arizona, currently on assignment at the Defense
Advanced Research Projects Agency. He is a Fellow of the
Association for the Advancement of Artificial Intelligence.

66 AI MAGAZINE